Abstract:

The integration of artificial intelligence (AI) and learning analytics into higher education presents new opportunities for monitoring student engagement and understanding of the course’s materials. It also enables ongoing support for open questions as the semester progresses in accordance with the syllabus topics when the goal is to support academic success. However, relying solely on summative assessments often limits instructors’ ability to intervene when students face persistent challenges early in the semester. While predictive analytics show promise, there remains a critical need to understand how combining traditional academic indicators with fine-grained traces from AI-assisted learning environments can help identify at-risk students before final evaluations.

This study investigated the predictive value of full academic support provided throughout the semester during computer-lab-classes and practices. These activities generated formative data that were used to forecast final exam success in an undergraduate technology course for engineering students. Drawing on data collected across a full academic semester, we integrate traditional performance metrics (such as ongoing semester grades, participation in structured course activities, and performance on formative pre-exam questions) with interaction telemetry derived from a syllabus-aware AI teaching assistant.

This study addresses two primary questions:
(1) How do early semester learning techniques and AI engagement patterns correlate with ultimate examination performance?
(2) How can the synthesis of these formative indicators be extended to reliably predict student success to enable timely pedagogical interventions during (and not after) the semester?

We employed a predictive modeling approach to evaluate student trajectories, framing the analysis within the context of authentic classroom integration rather than as a purely technical algorithmic optimization. Our conceptual modeling approach yielded a predictive accuracy of at least 60%, depending on the configuration of temporal indicators. Although the achieved accuracy was moderate, it reflected the intrinsic complexity and variability of student achievement. The integration of AI interaction logs with formative assessment data produces actionable indicators of progress, which allows for student support and solving difficulties that arise during the semester with a deeper understanding of the course’s materials and without increasing the teaching load on the course instructor. Crucially, this evaluation focuses on the practical implications for classroom workflows and early identification, rather than claiming definitive causal impacts on objective learning gains. The findings illustrate both the promise and the boundaries of using hybrid learning analytics for student support. This study provides practical recommendations for integrating AI-assisted learning traces with traditional academic performance, emphasizing the value of timely instructional intervention and syllabus-aligned support in higher education settings.

Keywords:

Predictive Modeling of Academic Success, Syllabus-Aware AI Teaching Assistant, AI-Assisted Learning Analytics, Early Identification of At-Risk Students.